Abstract
Coarse-grained reconfigurable arrays (CGRA) exhibit high performance, improved flexibility, low cost, and power efficiency for various application domains. Compute-intensive loop kernels are mapped to CGRA through modified modulo scheduling algorithms that integrate placement and routing. Most existing approaches are heavily influenced by VLIW compilation and FPGA synthesis techniques. A salient feature of these approaches is that data routing from a single source node to multiple destination nodes follow independent paths leading to resource wastage and hence inefficient schedule.We transform the CGRA mapping problem with route sharing into a graph minor problem. Our graph minor formalization provides a solid foundation for application mapping on CGRA. We provide an efficient framework based on graph mapping to solve this problem. Experimental validation shows that our approach leads to higher performance compared to state-of-the-art solutions with better resource utilization and minimal compilation time.